Claude Haiku 5.5
Key numbers
- Panel score
- 53rank 7 of 11
- Cost
- $1.32API list price
- Wall time
- 29 min
- Model time
- 22 min78% of wall time
- tok/s
- 214.7
- Tokens
- 12.3MOutput: 289k
- Cache hit
- 96%
- API calls
- 43
- Tool calls
- 44
- Renderer
- Intel UHD 630 (GPU)
- H1–H6
- Pass
- Cost per point
- $0.025
How the run ended
Finished with a passing video and an honest report. The cheapest and fastest of the Claude runs; the audio track is silent.
Blind panel
Scores of the three judges in every panel this video was part of.
| Panel | Letter | Anthropic | OpenAI | Median | Spread | Rank | |
|---|---|---|---|---|---|---|---|
| 14 Nov 2026 11 videos | J | 53 | 43 | 91 | 53 | 48 | 7 / 11 |
| 13 Nov 2026 10 videos | D | 53 | 41 | 58 | 53 | 17 | 7 / 10 |
| 11 Nov 2026 9 videos | G | 47 | 38 | 56 | 47 | 18 | 7 / 9 |
Sub-scores and the judges’ one-line notes from the panel of 14 Nov 2026.
| judge | Landmarks / 25 | City look / 25 | Camera & drone / 30 | Life / 10 | Overall / 10 | Total |
|---|---|---|---|---|---|---|
| Anthropic | 13 | 14 | 14 | 7 | 5 | 53 |
| OpenAI | 5 | 10 | 19 | 6 | 3 | 43 |
| 23 | 23 | 28 | 8 | 9 | 91 |
Judges’ notes
- AnthropicColorful streets and statued bridge, but upward-tilted and corridor-pressed frames and distorted perspectives hurt it.
- OpenAIThe visible drone and crowds cannot overcome generic windowed towers, distorted passages and weak landmark identity.
- GoogleStrongest all-around entry with recognisable St. Vitus, green-domed St. Nicholas, and Charles Bridge, rich baroque facades, and clean collision-free chase camera framing.
Time and speed
Wall time includes rendering and encoding. Model time is the sum of all model calls.
- Model generating
- Tools & waiting
| Wall time | 29 min |
|---|---|
| Model time | 22 min |
| Tools & waiting | 6.5 min |
| Model share of wall time | 78% |
| tok/s | 214.7 |
| Model calls | 43 |
Tokens and cost
Tokens summed over every model response. Tokens multiplied by the provider’s price.
| Uncached input | 88 |
|---|---|
| Cache read | 11,568,329 |
| Cache write | 467,721 |
| Output | 288,659 |
| Total | 12,324,797 |
| Cache-hit ratio | 96.1% |
| Cost | $1.32API list price |
| Output tokens per point | 5,446 |
| Points per $ | 40.2 |
Tools used
Tool calls by name, as the harness logged them. Names differ between harnesses.
Rendering
| Renderer | Intel UHD 630 (GPU) |
|---|---|
| Final render | 1.5 min |
| Per frame | – |
Hard requirements H1–H6
Checked independently from the delivered MP4 with ffprobe.
| H1 | h264 High yuv420p | Pass | |
|---|---|---|---|
| H1 | faststart | Pass | |
| H2 | 1080×1920 SAR1:1 | Pass | 1080x1920 SAR 1:1 |
| H3 | 30 fps | Pass | 30 fps |
| H4 | 55-60 s | Pass | 58.000 s |
| H5 | peak≤ 20Mbps, <200MB | Pass | peak 14.17 Mbps over any 1 s, 69.6 MB |
| H6 | AAC-LC 48k stereo | Pass |
From the agent’s report
Short, verbatim quotes from the REPORT.md the agent wrote itself.
Intro (0 to 2.2 s): St Vitus is not recognisable.
The spec asks for the drone visible with props at 0.5 s; this is not met.
Contact sheet and route map
